Information Security Electrical Engineering

Qiangsen He

*** ****, ******, *******, *** *P*

T 613-***-**** B ac1lg2@r.postjobfree.com

https://github.com/Summitofpyramid

Objective

Applying for the software development engineer position Summary

I am a newly graduated research-based master student in UOttawa. My study focuses on image processing, computer vision, machine learning, data mining and statistics.

{ Proficient in image and video processing with C++ and OpenCV

{ Skilled with machine learning including supervised, unsupervised learning and statistical analysis

{ Experienced with neural networks, gradient descent method, linear and nonlinear optimization

{ Skilled with data structure, operating system programming and object-oriented programming (OOP)

{ Solid knowledge in basic protocols, such as TCP/IP, MPLS, BGP, IGP, EIGRP

{ Experienced with and robotic kinematics and multithread programming

{ Experienced with time-series signal analysis (EEG spectrogram analysis) Skills

Languages: C, C++, Python, Linux shell scripting

Others: Data structure (C++ STL containers) and algorithms analysis, OpenCV, MySQL, Matlab,scikitLearn Education

Master of Applied Science Sep.2014-May.2014

School of Electrical Engineering and Computer Science, University of Ottawa, ON, Canada Bachelor of Engineering (GPA:3.78/4) Sep.2010-Jun.2014 School of Electronic Engineering, University of Electronic Science and Technology of China, Chengdu, China

Publications

Lili Pan, Qiangsen He, Yali Zheng, Mei Xie,Using Correlated Regression Models to Calculate Cumu- lative Attributes for Age Estimation, IEICE TRANSACTIONS on Information and Systems Vol.E98-D No.12 pp.2349-2352

Master Degree Thesis

Supervisor: Robert Laganiere

Description: Person re-identification (Re-ID) is to match an image of a person with other images of the same person. Most works try to use more discriminative features and metric learning to improve Re-ID accuracy. Statistical algorithms, machine learning and artificial neural networks have been exploited in Re-ID. In this thesis kernel local fisher discriminative analysis (KLFDA) and gradient descent optimization are combined together to learn a Mahalanobis distance metric. My detailed achievements are listed below.

{ Extracted the Gaussian of Gaussian (GOG) descriptors for given datasets

{ Reduced the dimensionality of GOG descriptor by KLFDA

{ Implemented a Mahalanobis metric based on gradient descent optimization

{ Improved the accuracy by maximally 5% on datasets including VIPeR, CUHK01, Prid_450s and GRID Research and Project Experience

Drowsy driving prediction Mar. 2017 - Present

Sleep Research Lab in University of Ottawa

Personal Role: Developer with Matlab

This project aims to predict drowsiness in driving based on EEG spectrogram analysis, eye tracking and driver’s responding time.

{ Devices configuration on PC and code review

{ EEG spectrogram analysis with Matlab, tracking eyes movement with Gazepoint Catch the professor (An android game) Sep.2014-Dec.2014 Personal Role: Developer with Java

{ Designed the algorithm for the app

{ Implemented the logic of the target’s moving strategy for the optimum moving routes Fast Food Restaurant Service Time Estimation Sep.2014-Sep.2016 advised by Professor Robert Laganiere, Lab of VIVA, University of Ottawa. Personal Role: Developer with C++

This co-operated project is aimed to estimate customers’ service time in the restaurant.The key problem is people re-identification. We extract customers’ feature descriptors and compare the probe image with gallery images, then a similarity score is computed.The whole project is realized by C++.

{ Estabalished the C++ framework of testing different descriptors on predefined datasets (VIPeR, PRID2011)

{ Implemented a new histogram-based descriptor with C++

{ Tested the performance of SIFT, LBP and MSCR descriptors on predefined datasets

{ Tested performance of descriptors on our own image dataset

{ Programmed with Python to plot the performance curve (CMC curve) of descriptors The Recognition of Moving Gestures Sep.2013-Jun.2014 Lab of Image Processing and Information Security, UESTC. Personal Role: Developer

This project is implemented with MATLAB, we use skeleton joints with depth data to describe a specific posture. To classify and recognize postures, a model of energy function for a single frame of consecutive motions is established. Mixture of CCA (Canonical Correlation Analysis) is used to train and predict the motions from CMU motion capture dataset.

{ Wrote the code to describe original poses of train data

{ Trained the original data and use them to predict test image’s pose Mixture of Regressors for Visual Mapping (sponsored by NSF) Apr.2013-Aug.2013 Lab of Image Processing and Information Security, UESTC. Personal Role: Matlab programmer

Age estimation is to learn the mapping function between input face image and target age,and this project is implemented with MATLAB. In this proposed method we extracted the GLOH (Gradient Location and Orientation Histogram) feature and reduce its dimensionality with LPP (Locality Preserving Pro- jection)method.Through imposing similarity regularization on related regressors, our proposed local linear regression model predicts the age robustly and accurately. Detailed achievements:

{ Implemented the trained pictures’ normalization of shape and illumination with AAM model

{ Extracted key GLOH features from normalized input

{ Proposed relatedness regularized mixture of regressors model

{ Evaluated the performance on FG_NET dataset

Contact this candidate